Booting from a SD card

GoogleGroups
1

I’ve been banging my head against a boot problem for a couple of days and have come to the conclusion that I need to ask someone.

Background: I have a bunch of BeagleBone Black Rev C boards. I want to load Gentoo onto a SD card. I used the “beaglebone black” SD creator script to do so, but pretty quickly came to realize that the script was making an “old style” SD image with a small VFAT partition and a ext4 partition. Looking at a current beaglebone SD image (which does boot just fine), I found the link to a page that explains how to create an image. So, I created an image using this script that I threw together based upon the instructions:

#!/bin/bash

some useful variables

SCRIPTNAME=$(echo $0 | sed ‘s|^.*/||g’)
DEVICE=""
MLOFILE=""
UBOOTIMAGE=""
ROOTFS=""

text parsing WON’T WORK unless it is in english

export LC_ALL=C

make sure there are enough options

if [ $# != 4 ]
then
echo “Usage:”
echo " $SCRIPTNAME <u-boot.img> "
echo " Example: $SCRIPTNAME /dev/sdc MLO u-boot.img nfs.tar.gz"
exit
fi

make sure the options are valid

if [[ -e $1 ]]
then
DEVICE=$1
else
echo “Invalid device!”
exit
fi

if [[ -e $2 ]]
then
MLOFILE=$2
else
echo “Incorrect MLO location!”
exit
fi

if [[ -e $3 ]]
then
UBOOTIMAGE=$3
else
echo “Incorrect u-boot.img location!”
exit
fi

if [[ -e $4 ]]
then
ROOTFS=$4
else
echo “Incorrect rootfs location!”
exit
fi

prompt for total destruction

echo “All data on “$1” now will be destroyed! Continue? [y/n]”
read ans
if ! [ $ans == ‘y’ ]
then
exit
fi

echo “[ clearing sectors…]”
dd if=/dev/zero of=$DEVICE bs=1M count=1

DISKSIZE=fdisk -l $DRIVE | grep Disk | awk '{print $5}'
echo " DISK SIZE: $DISKSIZE bytes"
echo

echo “[ writing MLO…]”
dd if=$MLOFILE of=$DEVICE count=1 seek=1 conv=notrunc bs=128K

echo “[ writing u-boot image…]”
dd if=$UBOOTIMAGE of=$DEVICE count=2 seek=1 conv=notrunc bs=384K

echo “[ creating root partition…]”
PARTITIONOFFSET=$(echo 38410243 | bc)
echo $PARTITIONOFFSET,0x83,- | sfdisk $DEVICE

echo “[ creating filesystem…]”
mkfs.ext4 -L rootfs -T small "$DEVICE"1 &> /dev/null

echo “[ copying files…]”
if ! [ -e /mnt/sdcard ]
then
mkdir /mnt/sdcard
if [ $? != 0 ]
then
echo “failed to create mountpoint!”
exit
fi
fi

mount "$DEVICE"1 /mnt/sdcard
if [ $? != 0 ]
then
echo “failed to mount filesystem!”
exit
fi
tar zxvf $ROOTFS -C /mnt/sdcard
chmod 755 /mnt/sdcard
umount "$DEVICE"1

echo “[Done]”

My image does not boot. There seem to be two issues:

  1. Watching the serial console during boot, I see “U-Boot SPL 2015.10-00001-g143c9ee (Nov 06 2015 - 15:27:19)
    bad magic”. Looking on this site, it seems like the U-Boot SPL should be picking up the MLO that I wrote to my disc and boot my new u-boot image, right? What would cause “bad magic”?

  2. When I interrupt u-boot and try to take a look at my ext4 partition, I see the following:

=> ext4ls mmc 0:1 /boot

1024 . 3145753 .. 102 fsf-funding.7 11 gpl.7 11619 gfdl.7

Even if the problem mentioned in #1, above, happens, the default u-boot should be able to boot my kernel as the directory mmc 0:1 /boot is populated with:

$ ls -la
total 10122
drwxrwxrwx 3 root root 1024 Jun 5 14:05 .
drwxrwxrwx 20 root root 1024 Jun 5 14:31 …
-rw-rw-rw- 1 root root 87161 Jun 2 14:11 am335x-boneblack.dtb
-rw-rw-rw- 1 root root 126411 Jun 2 14:12 .config
-rw-rw-rw- 1 root root 492 Jun 2 14:25 SOC.sh
-rw-rw-rw- 1 root root 2969405 Jun 2 14:11 System.map
drwxrwxrwx 2 root root 1024 May 31 12:56 uboot
-rw-rw-rw- 1 root root 405 Jun 5 14:05 uEnv.txt
-rwxrwxrwx 1 root root 7175152 Jun 2 14:10 vmlinuz-3.14.79

That’s what I see on my host system. u-boot seems to be listing files present in /usr/share/gcc-data/armv7a-hardfloat-linux-gnueabi/4.9.4/man/man7 rather than what is present in /boot. That would seem to indicate that u-boot is not reading the ext4 filesystem correctly. Is there some magic that needs to be present in an ext4 filesystem in order for u-boot to read it correctly? I tried shrinking my partition size down to 3.6GB to match the beaglebone SD image, but that had no effect. I didn’t see anything about directory depth limits nor about size limits, but clearly something is not working properly.

Any answers or pointers to better (correct) documentation are greatly appreciated.